These local (smaller) LLMs are good for:
>Flavor text
>Pointless dialogues
>Generating descriptions when details are provided

but they suck dick at everything else, you can't get any consistent output from them, no matter how much you spoon feed or tweak them. They're not reliable for decision making or analyzing emotions or mental states, because of the stupid positivity bias.
Because of this, you also can’t use them as logical engines